DScribe: Library of descriptors for machine learning in materials science

Computer Physics Communications, Journal Year: 2019, Volume and Issue: 247, P. 106949 - 106949

Published: Sept. 26, 2019

DScribe is a software package for machine learning that provides popular feature transformations ("descriptors") atomistic materials simulations. accelerates the application of property prediction by providing user-friendly, off-the-shelf descriptor implementations. The currently contains implementations Coulomb matrix, Ewald sum sine Many-body Tensor Representation (MBTR), Atom-centered Symmetry Function (ACSF) and Smooth Overlap Atomic Positions (SOAP). Usage illustrated two different applications: formation energy solids ionic charge atoms in organic molecules. freely available under open-source Apache License 2.0.

Language: Английский

---

Big-Data Science in Porous Materials: Materials Genomics and Machine Learning

Chemical Reviews, Journal Year: 2020, Volume and Issue: 120(16), P. 8066 - 8129

Published: June 10, 2020

By combining metal nodes with organic linkers we can potentially synthesize millions of possible frameworks (MOFs). At present, have libraries over ten thousand synthesized materials and in-silico predicted materials. The fact that so many opens exciting avenues to tailor make a material is optimal for given application. However, from an experimental computational point view simply too screen using brute-force techniques. In this review, show having allows us use big-data methods as powerful technique study these discover complex correlations. first part the review gives introduction principles science. We emphasize importance data collection, augment small sets, how select appropriate training sets. An important are different approaches used represent in feature space. also includes general overview ML techniques, but most applications porous supervised our focused on ML. particular, method optimize process quantify performance methods. second part, been applied discuss field gas storage separation, stability materials, their electronic properties, synthesis. range topics illustrates large variety be studied Given increasing interest scientific community ML, expect list rapidly expand coming years.

Language: Английский

---

Representations of Materials for Machine Learning

Annual Review of Materials Research, Journal Year: 2023, Volume and Issue: 53(1), P. 399 - 426

Published: April 18, 2023

High-throughput data generation methods and machine learning (ML) algorithms have given rise to a new era of computational materials science by the relations between composition, structure, properties exploiting such for design. However, build these connections, must be translated into numerical form, called representation, that can processed an ML model. Data sets in vary format (ranging from images spectra), size, fidelity. Predictive models scope interest. Here, we review context-dependent strategies constructing representations enable use as inputs or outputs models. Furthermore, discuss how modern techniques learn transfer chemical physical information tasks. Finally, outline high-impact questions not been fully resolved thus require further investigation.

Language: Английский

---

Machine Learning Descriptors for Data‐Driven Catalysis Study

Advanced Science, Journal Year: 2023, Volume and Issue: 10(22)

Published: May 16, 2023

Traditional trial-and-error experiments and theoretical simulations have difficulty optimizing catalytic processes developing new, better-performing catalysts. Machine learning (ML) provides a promising approach for accelerating catalysis research due to its powerful predictive abilities. The selection of appropriate input features (descriptors) plays decisive role in improving the accuracy ML models uncovering key factors that influence activity selectivity. This review introduces tactics utilization extraction descriptors ML-assisted experimental research. In addition effectiveness advantages various descriptors, their limitations are also discussed. Highlighted both 1) newly developed spectral performance prediction 2) novel paradigm combining computational through suitable intermediate descriptors. Current challenges future perspectives on application techniques presented.

Language: Английский

---

How to train a neural network potential

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 159(12)

Published: Sept. 27, 2023

The introduction of modern Machine Learning Potentials (MLPs) has led to a paradigm change in the development potential energy surfaces for atomistic simulations. By providing efficient access energies and forces, they allow us perform large-scale simulations extended systems, which are not directly accessible by demanding first-principles methods. In these simulations, MLPs can reach accuracy electronic structure calculations, provided that have been properly trained validated using suitable set reference data. Due their highly flexible functional form, construction be done with great care. this Tutorial, we describe necessary key steps training reliable MLPs, from data generation via final validation. procedure, is illustrated example high-dimensional neural network potential, general applicable many types MLPs.

Language: Английский

---

Updates to the DScribe library: New descriptors and derivatives

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 158(23)

Published: June 20, 2023

We present an update of the DScribe package, a Python library for atomistic descriptors. The extends DScribe's descriptor selection with Valle-Oganov materials fingerprint and provides derivatives to enable more advanced machine learning tasks, such as force prediction structure optimization. For all descriptors, numeric are now available in DScribe. many-body tensor representation (MBTR) Smooth Overlap Atomic Positions (SOAP), we have also implemented analytic derivatives. demonstrate effectiveness models Cu clusters perovskite alloys.

Language: Английский

---

Machine learning interatomic potential: Bridge the gap between small-scale models and realistic device-scale simulations

iScience, Journal Year: 2024, Volume and Issue: 27(5), P. 109673 - 109673

Published: April 4, 2024

Machine learning interatomic potential (MLIP) overcomes the challenges of high computational costs in density-functional theory and relatively low accuracy classical large-scale molecular dynamics, facilitating more efficient precise simulations materials research design. In this review, current state four essential stages MLIP is discussed, including data generation methods, material structure descriptors, six unique machine algorithms, available software. Furthermore, applications various fields are investigated, notably phase-change memory materials, searching, properties predicting, pre-trained universal models. Eventually, future perspectives, consisting standard datasets, transferability, generalization, trade-off between complexity MLIPs, reported.

Language: Английский

---

Transition metal-anchored BN tubes as single-atom catalysts for NO reduction reaction: A study of DFT and deep learning

Fuel, Journal Year: 2025, Volume and Issue: 386, P. 134302 - 134302

Published: Jan. 7, 2025

Language: Английский

---

Optimizing many-body atomic descriptors for enhanced computational performance of machine learning based interatomic potentials

Physical review. B./Physical review. B, Journal Year: 2019, Volume and Issue: 100(2)

Published: July 30, 2019

We explore different ways to simplify the evaluation of smooth overlap atomic positions (SOAP) many-body descriptor [Bart\'{o}k et al., Phys. Rev. B 87, 184115 (2013)]. Our aim is improve computational efficiency SOAP-based similarity kernel construction. While these improved descriptors can be used for general characterization and interpolation properties, their main target application accelerated machine-learning-based interatomic potentials within Gaussian approximation potential (GAP) framework Lett. 104, 136403 (2010)]. achieve this objective by expressing densities in an approximate separable form, which decouples radial angular channels. then express elements SOAP (i.e., expansion coefficients densities) analytical form given a particular choice basis set. Finally, we derive recursion formulas coefficients. This new allows tenfold speedups compared previous implementations, while improving stability distant neighbors, without degradation power GAP models.

Language: Английский

---

Band Gap Prediction for Large Organic Crystal Structures with Machine Learning

Advanced Quantum Technologies, Journal Year: 2019, Volume and Issue: 2(7-8)

Published: July 1, 2019

Machine-learning models are capable of capturing the structure-property relationship from a dataset computationally demanding ab initio calculations. Over past two years, Organic Materials Database (OMDB) has hosted growing number calculated electronic properties previously synthesized organic crystal structures. The complexity crystals contained within OMDB, which have on average 82 atoms per unit cell, makes this database challenging platform for machine learning applications. In paper, focus is predicting band gap represents one basic crystalline materials. With aim, consistent 12 500 structures and their corresponding DFT released, freely available download at https://omdb.mathub.io/dataset. An ensemble state-of-the-art reach mean absolute error (MAE) 0.388 eV, corresponds to percentage 13% an 3.05 eV. Finally, trained employed predict 260 092 materials Crystallography Open (COD) made online so that predictions can be obtained any arbitrary structure uploaded by user.

Language: Английский

---